LaTtE-Flow partitions transformer layers into timestep-specific groups for flow matching, activating only one group per sampling step to speed up image generation in unified multimodal models.
Deep compression autoencoder for efficient high-resolution diffusion models
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
LaTtE-Flow: Layerwise Timestep-Expert Flow-based Transformer
LaTtE-Flow partitions transformer layers into timestep-specific groups for flow matching, activating only one group per sampling step to speed up image generation in unified multimodal models.